Breathing state evaluation method and system based on optical blood flow phase signals
By collecting and processing PPG data based on optical blood flow phase signals, extracting multi-dimensional feature parameters, and using SVM model training and optimization, the accuracy and information burden problems of respiratory state assessment in existing technologies are solved, and a respiratory state assessment that is closer to reality is achieved.
Patent Information
- Application Number
- CN202510890729.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-06-30
- Publication Date
- 2025-10-17
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
In the existing technology, PPG-based respiratory status assessment methods rely on single amplitude modulation or frequency modulation, which makes it difficult to accurately identify complex respiratory patterns, and the excessive number of physiological parameters leads to information overload.
A method based on optical blood flow phase signals was used to collect PPG data under natural breathing conditions, extract amplitude, frequency and nonlinear characteristic parameters, and train and optimize the SVM classifier model to construct a respiratory state assessment model.
It achieves more accurate respiratory status assessment, overcomes the limitations of a single indicator, adapts to different scenario needs, and supports doctors in tracing the dominant factors causing score decline.
Smart Images
Figure CN120804854A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of respiratory state evaluation of PPG signal, in particular to a respiratory state evaluation method and system based on optical blood flow signal. BACKGROUND
[0002] In the prior art, the evaluation of respiratory state is usually based on the ventilation parameters on the ventilator or anesthetic machine. The respiratory state evaluation based entirely on physiological parameters has not been seen.
[0003] In recent years, the research on PPG-based respiratory monitoring has made some progress, but there are still the following problems:
[0004] Single respiratory feature extraction algorithm: existing methods mostly rely on single analysis of amplitude modulation (AM) or frequency modulation (FM), which is difficult to accurately identify complex respiratory patterns;
[0005] Various physiological parameters: these signals and parameters provide rich parameter index information for clinical practice, which to some extent represents part of the physiological state. Too many physiological parameter signals cause information burden. SUMMARY
[0006] In view of the deficiencies of the prior art, the present application provides a respiratory state evaluation method and system based on optical blood flow signal to solve the problems in the background art.
[0007] To achieve the above purpose, the present application realizes the following technical scheme: a respiratory state evaluation method based on optical blood flow signal, comprising the following steps:
[0008] Collecting PPG data under natural respiratory state and processing to obtain natural PGG data set;
[0009] Extracting feature PPG data set of target respiratory behavior in the natural PGG data set;
[0010] Extracting respiratory feature parameters in the feature PPG data set and calculating to obtain respiratory parameters;
[0011] Dividing the respiratory parameters into a training set and a validation set, inputting the training set into a SVM classifier model, training the SVM classifier model, and obtaining a respiratory state evaluation model when the training progress of the SVM classifier model meets a preset training termination condition;
[0012] Inputting the validation set into the respiratory state evaluation model for verification, and optimizing the respiratory state evaluation model based on the verification result;
[0013] Obtain the PPG data to be evaluated, and input the optimized respiratory state evaluation model to obtain the respiratory state evaluation result.
[0014] Preferably, the respiratory feature parameters include amplitude features, frequency features, and nonlinear features.
[0015] The calculation formula of the amplitude feature is:
[0016] ;
[0017] ;
[0018] Wherein, PPG is the PPG signal, p.v. is the Cauchy principal value integral, τ is the integral variable, and E(t) is the envelope signal.
[0019] ;
[0020] Wherein, BR represents the respiratory rate, P represents the number of peak values of E(t) in the time window, and D represents the time (s) of the analysis window.
[0021] Preferably, the calculation formula of the frequency feature is:
[0022] ;
[0023] Wherein, The respiratory frequency (Hz) is represented by f, The frequency corresponding to the maximum spectral energy in the range of 0.1-0.5 Hz is searched.
[0024] Preferably, the calculation formula of the nonlinear feature is:
[0025] ;
[0026] Wherein, m represents the embedding dimension, r represents the similarity threshold, and N represents the signal length, The logarithm of the template pair satisfying the similarity condition in the m+1-dimensional space is represented by N, The logarithm of the template pair satisfying the similarity condition in the m-dimensional space is represented by N.
[0027] Preferably, the composite score function of the respiratory state model for the respiratory state is:
[0028] ;
[0029] Wherein, The weighted F1 score is represented by F1, the Matthews correlation coefficient is represented by MCC, the area under the ROC curve is represented by AUC, and α, β, and γ are all dynamic weight coefficients, and α+β+γ=1.
[0030] The calculation formula of the weighted F1-socre is:
[0031] ;
[0032] k represents the number of categories, ni represents the number of samples of the ith category, and n represents the total number of samples, represents the F1-socre of the ith category;
[0033] The calculation formula of the Matthews correlation coefficient is:
[0034] ;
[0035] where TP represents true positive, TN represents true negative, FP represents false positive, and FN represents false negative;
[0036] The calculation formula of the ROC-AUC is:
[0037] ;
[0038] where TPR represents the true positive rate: , and FPR represents the false positive rate: .
[0039] Preferably, a respiration state evaluation system based on an optical blood flow belief signal is used to implement the respiration state evaluation method based on the optical blood flow belief signal, and comprises:
[0040] A data acquisition module is configured to acquire PPG data in a natural respiration state and process the PPG data to obtain a natural PGG data set;
[0041] A data processing module is configured to extract a feature PPG data set of a target respiration behavior in the natural PGG data set, and further configured to extract a respiration feature parameter in the feature PPG data set and calculate the respiration feature parameter to obtain a respiration parameter;
[0042] A model construction module is configured to divide the respiration parameter into a training set and a verification set, input the training set into an SVM classifier model, train the SVM classifier model, and obtain a respiration state evaluation model when a training progress of the SVM classifier model meets a preset training termination condition;
[0043] The model construction module is further configured to input the verification set into the respiration state evaluation model for verification, and optimize the respiration state evaluation model based on a verification result;
[0044] A state evaluation module is configured to acquire PPG data to be evaluated and input the PPG data to be evaluated into the optimized respiration state evaluation model to obtain a respiration state evaluation result.
[0045] Preferably, the respiratory feature parameters include amplitude features, frequency features and non-linear features.
[0046] The calculation formula of the amplitude features is:
[0047] ;
[0048] ;
[0049] wherein PPG is the PPG signal, p.v. is the Cauchy principal value integral, τ is the integral variable, and E(t) is the envelope signal;
[0050] ;
[0051] wherein BR represents the respiratory rate, P represents the number of peak values of E(t) within the time window, and D represents the time (s) of the analysis window.
[0052] Preferably, the calculation formula of the frequency features is:
[0053] ;
[0054] wherein represents the respiratory frequency (Hz), the frequency corresponding to the maximum spectral energy in the range of 0.1-0.5 Hz is searched.
[0055] Preferably, the calculation formula of the non-linear features is:
[0056] ;
[0057] wherein m represents the embedding dimension, r represents the similarity threshold, and N represents the signal length, represents the logarithm of the number of template pairs satisfying the similarity condition in the m+1-dimensional space, represents the logarithm of the number of template pairs satisfying the similarity condition in the m-dimensional space.
[0058] Preferably, the composite score function of the respiratory state model for the respiratory state is:
[0059] ;
[0060] wherein represents the weighted F1 score, MCC represents the Matthew correlation coefficient, AUC represents the area under the ROC curve, and α, β and γ are dynamic weight coefficients, and α+β+γ=1;
[0061] The calculation formula of the weighted F1-socre is:
[0062] ;
[0063] k represents the number of categories, ni represents the number of samples of the ith category, and n represents the total number of samples, represents the F1-score of the ith category;
[0064] The calculation formula of the Matthews correlation coefficient is:
[0065] ;
[0066] Where TP represents true positive, TN represents true negative, FP represents false positive, and FN represents false negative;
[0067] The calculation formula of the ROC-AUC is:
[0068] ;
[0069] Where TPR represents the true positive rate: , and FPR represents the false positive rate: .
[0070] The application provides a respiration state evaluation method and system based on an optical blood flow belief signal, which has the following beneficial effects: PPG data in a natural respiration state is collected and processed to obtain a natural PGG data set; feature PPG data set of target respiration behavior in the natural PGG data set is extracted; respiration feature parameters in the feature PPG data set are extracted and calculated to obtain respiration parameters; the respiration parameters are divided into a training set and a verification set, the training set is input into an SVM classifier model, the SVM classifier model is trained, when the training progress of the SVM classifier model meets a preset training termination condition, a respiration state evaluation model is obtained; the verification set is input into the respiration state evaluation model for verification, the respiration state evaluation model is optimized based on the verification result; PPG data to be evaluated is obtained and input into the optimized respiration state evaluation model to obtain a respiration state evaluation result, that is, the application can more closely evaluate the respiration state. BRIEF DESCRIPTION OF DRAWINGS
[0071] Figure 1 The application provides a respiration state evaluation method and system based on an optical blood flow belief signal, which has the following beneficial effects: PPG data in a natural respiration state is collected and processed to obtain a natural PGG data set; feature PPG data set of target respiration behavior in the natural PGG data set is extracted; respiration feature parameters in the feature PPG data set are extracted and calculated to obtain respiration parameters; the respiration parameters are divided into a training set and a verification set, the training set is input into an SVM classifier model, the SVM classifier model is trained, when the training progress of the SVM classifier model meets a preset training termination condition, a respiration state evaluation model is obtained; the verification set is input into the respiration state evaluation model for verification, the respiration state evaluation model is optimized based on the verification result; PPG data to be evaluated is obtained and input into the optimized respiration state evaluation model to obtain a respiration state evaluation result, that is, the application can more closely evaluate the respiration state.
[0072] Figure 2 The application provides a respiration state evaluation method and system based on an optical blood flow belief signal, which has the following beneficial effects: PPG data in a natural respiration state is collected and processed to obtain a natural PGG data set; feature PPG data set of target respiration behavior in the natural PGG data set is extracted; respiration feature parameters in the feature PPG data set are extracted and calculated to obtain respiration parameters; the respiration parameters are divided into a training set and a verification set, the training set is input into an SVM classifier model, the SVM classifier model is trained, when the training progress of the SVM classifier model meets a preset training termination condition, a respiration state evaluation model is obtained; the verification set is input into the respiration state evaluation model for verification, the respiration state evaluation model is optimized based on the verification result; PPG data to be evaluated is obtained and input into the optimized respiration state evaluation model to obtain a respiration state evaluation result, that is, the application can more closely evaluate the respiration state. DETAILED DESCRIPTION
[0073] Embodiments of the present application are described in detail below with reference to the attached drawing figures, wherein the same or like component having the same or similar designations or functions throughout the figures are denoted with the same or like designations. The embodiments described below are presented by way of example only and are not intended to limit the present application as defined by the appended claims and their equivalents.
[0074] The disclosure given below provides many different embodiments or examples for implementing different structures of the present application. For the purpose of simplification of the present application disclosure, the components and settings of specific examples are described below. Of course, they are only examples and the purpose is not to limit the present application. In addition, the present application can repeatedly refer to numbers and / or letters in different examples, and such repetition is for the purpose of simplification and clarity, which does not indicate the relationship between the various embodiments and / or settings discussed. In addition, the present application provides examples of various specific processes and materials, but those skilled in the art can realize the application of other processes and / or the use of other materials.
[0075] As shown in Figure 1 , the embodiment of the present application provides a method for evaluating the respiratory state based on the optical blood flow signal, comprising the following steps:
[0076] S1: Collecting PPG data in natural breathing state and processing to obtain natural PGG data set;
[0077] Wherein, the PPG data is collected by dual-wavelength PPG sensor (red light 660nm + infrared light 940nm), sampling rate ≥100Hz, ADC resolution ≥12bit.
[0078] Specifically, dual-wavelength can distinguish between superficial and deep blood flow, and high sampling rate meets Nyquist theorem, so that the data accuracy is higher.
[0079] It should be noted that the processing process includes band-pass filtering, motion artifact elimination and baseline correction, and LMS algorithm can be used to eliminate running artifact. LMS algorithm is a prior art means, which is not described in detail here.
[0080] S2: Extracting the feature PPG data set of the target respiratory behavior in the natural PGG data set;
[0081] S3: Extracting the respiratory feature parameters in the feature PPG data set and calculating to obtain the respiratory parameters;
[0082] Specifically, the respiratory feature parameters include amplitude feature, frequency feature and nonlinear feature.
[0083] The calculation formula of amplitude feature is:
[0084] ;
[0085] ;
[0086] wherein PPG is the PPG signal, p.v. is the Cauchy principal value integral, tau is the integral variable, and E(t) is the envelope signal;
[0087] ;
[0088] wherein BR represents the breathing rate, P represents the number of peaks of E(t) within a time window, and D represents the time of the analysis window (s).
[0089] The formula for calculating the frequency feature is:
[0090] ;
[0091] wherein represents the breathing frequency (Hz), The frequency corresponding to the maximum spectral energy in the range of 0.1-0.5 Hz is searched.
[0092] The formula for calculating the nonlinear feature is:
[0093] ;
[0094] wherein m represents the embedding dimension, r represents the similarity threshold, and N represents the signal length, represents the logarithm of the number of template pairs that satisfy the similarity condition in the m+1-dimensional space, represents the logarithm of the number of template pairs that satisfy the similarity condition in the m-dimensional space.
[0095] It should be noted that the respiratory feature parameters also include the inhalation / expiration time ratio.
[0096] Specifically, by comprehensively analyzing the multi-dimensional data of the respiratory features, the respiratory state can be more accurately determined.
[0097] S4: dividing the respiratory parameters into a training set and a validation set, inputting the training set into an SVM classifier model, training the SVM classifier model, and obtaining a respiratory state evaluation model when the training progress of the SVM classifier model meets a preset training termination condition;
[0098] It should be noted that SVM (support vector machine) is a binary classification model, and its basic model is a linear classifier with maximum interval defined in a feature space, which uses maximum interval to find the optimal separation hyperplane. SVM theory provides a way to avoid the complexity of high-dimensional space, and directly uses the inner product function (kernel function) of this space to solve the decision problem in the corresponding high-dimensional space. When the kernel function is known, the difficulty of solving the high-dimensional space problem can be simplified. At the same time, SVM is based on small sample statistical theory, which meets the purpose of machine learning. Moreover, support vector machine has better generalization ability than neural network.
[0099] Specifically, the composite score function of the respiratory state model for the respiratory state is:
[0100] ;
[0101] Wherein, MCC represents the Matthews correlation coefficient, AUC represents the area under the ROC curve, and α, β and γ are dynamic weight coefficients, and α+β+γ=1.
[0102] The calculation formula of the weighted F1-score is:
[0103] ;
[0104] k represents the number of categories, ni represents the number of samples of the ith category, and n represents the total number of samples, represents the F1-score of the ith category;
[0105] The calculation formula of the Matthews correlation coefficient is:
[0106] ;
[0107] Wherein, TP represents true positive, TN represents true negative, FP represents false positive, and FN represents false negative.
[0108] The calculation formula of ROC-AUC is:
[0109] ;
[0110] Wherein, TPR represents the true positive rate: , and FPR represents the false positive rate: .
[0111] It should be noted that through multi-dimensional fusion, the limitations of single index are overcome.
[0112] Scene adaptation: adapt to different scene requirements such as ICU, home monitoring through weight adjustment
[0113] Strong interpretability: CS decomposition function supports doctors to trace the leading factors of score reduction
[0114] S5: input the verification set into the respiratory state evaluation model for verification, and optimize the respiratory state evaluation model based on the verification result;
[0115] S6: obtain the PPG data to be evaluated, and input it into the optimized respiratory state evaluation model to obtain the respiratory state evaluation result.
[0116] The respiratory state evaluation method based on optical blood flow belief signal provided in the embodiment, by collecting PPG data under natural breathing state and processing, natural PGG data set is obtained; the feature PPG data set of the target respiratory behavior in the natural PGG data set is extracted; the respiratory feature parameters in the feature PPG data set are extracted and calculated, and the respiratory parameters are obtained; the respiratory parameters are divided into training set and verification set, the training set is input into the SVM classifier model, the SVM classifier model is trained, when the training progress of the SVM classifier model meets the preset training termination condition, the respiratory state evaluation model is obtained; the verification set is input into the respiratory state evaluation model for verification, and the respiratory state evaluation model is optimized based on the verification result; the PPG data to be evaluated is obtained, and it is input into the optimized respiratory state evaluation model, and the respiratory state evaluation result is obtained, that is, the respiratory state can be evaluated more close to the truth.
[0117] A respiratory state evaluation system based on optical blood flow belief signal, for realizing the respiratory state evaluation method based on optical blood flow belief signal described above, comprising:
[0118] The data acquisition module is used for collecting PPG data under natural breathing state and processing, and obtaining natural PGG data set;
[0119] The data processing module is used for extracting the feature PPG data set of the target respiratory behavior in the natural PGG data set; and is also used for extracting the respiratory feature parameters in the feature PPG data set and calculating, and obtaining the respiratory parameters;
[0120] The model construction module is used for dividing the respiratory parameters into training set and verification set, inputting the training set into the SVM classifier model, training the SVM classifier model, and obtaining the respiratory state evaluation model when the training progress of the SVM classifier model meets the preset training termination condition;
[0121] The model construction module is also used for inputting the verification set into the respiratory state evaluation model for verification, and optimizing the respiratory state evaluation model based on the verification result;
[0122] The state evaluation module acquires the PPG data to be evaluated and inputs the PPG data into the optimized respiratory state evaluation model to obtain a respiratory state evaluation result.
[0123] Further, the respiratory feature parameters include amplitude features, frequency features and nonlinear features.
[0124] The calculation formula of the amplitude features is:
[0125] ;
[0126] ;
[0127] wherein, PPG is a PPG signal, p.v. is a Cauchy principal value integral, τ is an integral variable, and E(t) is an envelope signal;
[0128] ;
[0129] wherein, BR represents a respiratory rate, P represents a peak value number of E(t) in a time window, and D represents a time (s) of an analysis window.
[0130] Further, the calculation formula of the frequency features is:
[0131] ;
[0132] wherein, represents a respiratory frequency (Hz), a frequency corresponding to a maximum spectral energy in a range of 0.1-0.5 Hz is searched.
[0133] Further, the calculation formula of the nonlinear features is:
[0134] ;
[0135] wherein, m represents an embedding dimension, r represents a similarity threshold, and N represents a signal length, represents a logarithm of a template pair satisfying a similarity condition in an m+1-dimensional space, represents a logarithm of a template pair satisfying a similarity condition in an m-dimensional space.
[0136] Further, a composite score function of the respiratory state model for a respiratory state is:
[0137] ;
[0138] wherein, represents a weighted F1 score, MCC represents a Matthews correlation coefficient, AUC represents an area under an ROC curve, and α, β and γ are all dynamic weight coefficients, and α+β+γ=1.
[0139] The calculation formula of the weighted F1-socre is:
[0140] ;
[0141] k represents the number of categories, ni represents the number of samples of the ith category, and n represents the total number of samples, represents the F1-socre of the ith category;
[0142] The calculation formula of the Matthews correlation coefficient is:
[0143] ;
[0144] wherein TP represents true positive, TN represents true negative, FP represents false positive, and FN represents false negative;
[0145] The calculation formula of the ROC-AUC is:
[0146] ;
[0147] wherein TPR represents the true positive rate: , and FPR represents the false positive rate: ,
[0148] Although the embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A respiratory state assessment method based on optical blood flow phase signal, characterized in that: The following steps are involved: Collect and process PPG data under natural breathing conditions to obtain a natural PGG dataset; Extract the characteristic PPG dataset of the target breathing behavior in the natural PGG dataset; Extract the respiratory characteristic parameters from the feature PPG data set and calculate them to obtain the respiratory parameters; Dividing respiratory parameters into a training set and a validation set, inputting the training set into an SVM classifier model, training the SVM classifier model, and obtaining a respiratory state assessment model when the training progress of the SVM classifier model meets a preset training termination condition; The validation set is input into the respiratory state assessment model for validation, and the respiratory state assessment model is optimized based on the validation results; The PPG data to be evaluated is obtained and input into the optimized respiratory state evaluation model to obtain the respiratory state evaluation result.
2. A respiratory state assessment method based on optical blood flow phase signal according to claim 1, characterized in that: The respiratory characteristic parameters include amplitude characteristics, frequency characteristics and nonlinear characteristics; The calculation formula of the amplitude characteristic is: ； ； Where PPG is the PPG signal, pv is the Cauchy principal value integral, τ is the integral variable, and E(t) is the envelope signal; ； Where BR represents the respiratory rate, P represents the number of peaks of E(t) within the time window, and D represents the time of the analysis window (s).
3. The respiratory state assessment method based on optical blood flow phase signal according to claim 2, characterized in that: The calculation formula of the frequency characteristic is: ； in, represents the respiratory rate (Hz), Find the frequency corresponding to the maximum spectrum energy in the range of 0.1-0.5Hz.
4. The respiratory state assessment method based on optical blood flow phase signal according to claim 3, characterized in that: The calculation formula of the nonlinear characteristics is: ； Among them, m represents the embedding dimension, r represents the similarity threshold, and N represents the signal length. represents the number of template pairs that meet similarity conditions in the m+1 dimensional space, Represents the number of template pairs that meet the similarity condition in m-dimensional space.
5. The respiratory state assessment method based on optical blood flow phase signal according to claim 1, characterized in that: The composite scoring function of the respiratory state model for the respiratory state is: ； in, represents the weighted F1 score, MCC represents the Matthews correlation coefficient, AUC represents the area under the ROC curve, α, β, and γ are all dynamic weight coefficients, and α+β+γ=1; The calculation formula of weighted F1-score is: ； k represents the number of categories, ni represents the number of samples in category i, and n represents the total number of samples. represents the i-th class F1-socre; The calculation formula of Matthews correlation coefficient is: ； Among them, TP represents true positive, TN represents true negative, FP represents false positive, and FN represents false negative; The calculation formula for ROC-AUC is: ； Where TPR represents the true positive rate: , FPR represents the false positive rate: .
6. A respiratory state assessment system based on optical blood flow phase signals, used to implement the respiratory state assessment method based on optical blood flow phase signals according to any one of claims 1 to 5, characterized in that: include: The data acquisition module is used to collect and process PPG data under natural breathing conditions to obtain a natural PGG data set; The data processing module is used to extract the characteristic PPG dataset of the target breathing behavior in the natural PGG dataset; It is also used to extract and calculate the respiratory characteristic parameters in the feature PPG data set to obtain the respiratory parameters; A model construction module is used to divide the respiratory parameters into a training set and a validation set, input the training set into an SVM classifier model, train the SVM classifier model, and obtain a respiratory state assessment model when the training progress of the SVM classifier model meets a preset training termination condition; The model building module is further used to input the validation set into the respiratory state assessment model for validation, and optimize the respiratory state assessment model based on the validation results; The state assessment module obtains the PPG data to be evaluated and inputs it into the optimized respiratory state assessment model to obtain the respiratory state assessment result.
7. A respiratory state assessment system based on optical blood flow phase signals according to claim 6, characterized in that: The respiratory characteristic parameters include amplitude characteristics, frequency characteristics and nonlinear characteristics; The calculation formula of the amplitude characteristic is: ； ； Where PPG is the PPG signal, pv is the Cauchy principal value integral, τ is the integral variable, and E(t) is the envelope signal; ； Where BR represents the respiratory rate, P represents the number of peaks of E(t) within the time window, and D represents the time of the analysis window (s).
8. The respiratory state assessment system based on optical blood flow phase signal according to claim 7, characterized in that: The calculation formula of the frequency characteristic is: ； in, represents the respiratory rate (Hz), Find the frequency corresponding to the maximum spectrum energy in the range of 0.1-0.5Hz.
9. The respiratory state assessment system based on optical blood flow phase signal according to claim 8, characterized in that: The calculation formula of the nonlinear characteristics is: ； Among them, m represents the embedding dimension, r represents the similarity threshold, and N represents the signal length. represents the number of template pairs that meet similarity conditions in the m+1 dimensional space, Represents the number of template pairs that meet the similarity condition in m-dimensional space.
10. The respiratory state assessment system based on optical blood flow phase signal according to claim 9, characterized in that: The composite scoring function of the respiratory state model for the respiratory state is: ； in, represents the weighted F1 score, MCC represents the Matthews correlation coefficient, AUC represents the area under the ROC curve, α, β, and γ are all dynamic weight coefficients, and α+β+γ=1; The calculation formula of weighted F1-score is: ； k represents the number of categories, ni represents the number of samples in category i, and n represents the total number of samples. represents the i-th class F1-socre; The calculation formula of Matthews correlation coefficient is: ； Among them, TP represents true positive, TN represents true negative, FP represents false positive, and FN represents false negative; The calculation formula for ROC-AUC is: ； Where TPR represents the true positive rate: , FPR represents the false positive rate: .